Camera Height vs Volume Width Study

Study Overview

The objective of this study was to investigate the affect of camera height as a function of volume width on lower limb kinematics during normal walking using Theia3D. Video data was recorded for 9 subjects using 32 Sony RX0II cameras and processed in Theia3D (v2023.1.0.3161p9, model v10.0.0, 3DOF knee, 8 Hz cutoff frequency).

Camera Heights and Volume widths

Camera Heights
The 32 cameras were placed in a standard ellipse-like configuration (see Figure 1). Cameras were placed at 4 heights for each camera location:

• Group 1: 0.85m (hip height)
• Group 2: 1.65m (shoulder height)
• Group 3: 2.45m (HMRL standard setup)
• Group 4: 3.30m (HMRL scaffold/ceiling)

Group 0 included all 32 cameras during processing.

Volume Width
For each width the cameras were moved and re-aimed to get the subject in all camera views as much as possible given the width constraints.

Width 1 was the standard HMRL volume length (~13.8m Long) and width. Width 2 and 3 were narrower. With width 3 trying to simulate hallway.

• Width 1: 6.9 m
• Width 2: 5.5 m
• Width 3: 3.3 m (not added yet)

A 5th and 6th group were added to Width 3 since the sagittal cameras for groups 1 and 2 could not get the full subjects. The sagittal cameras were replaced with videos from group 3.

• Group 51: is the same as group1 for width 3 but the sagittal ameras are replaced with cameras from group 3 (HMRL standard setup).
• Group 62: is the same as group2 for width 3 but the sagittal cameras are replaced with cameras from group 3 (HMRL standard setup).

Calibration
A single calibration was performed for each width with all 32 cameras. Separate calibration files were then built by extracting the cameras for each group

Data Processing
Only steps within a regional of interest were kept for analysis. The region was defined to only include steps where the subject was visible by all cameras. The same region was applied to each grouping to ensure the same steps were used for analysis. The number of steps included reduced as a function of width. Both legs were included and averaged.

Average plots by camera height and volume width

The plots below are interactive:

• Click and drag to zoom.
• Hover over the line to see the group/width.

Draft interpretations per Jereme and Rob:

• Height doesn’t look to cause large differences, at least at these two widths (narrowest width still processing).
• The largest differences exist between the lowest and the highest heights as expected.
• The data when 36 cameras are included falls in between as expected, and biased towards the groupings that share similar views (Groups 2 and 3, i.e. shoulder height/above head height).
• There doesn’t seem to be a specific, consistent effect of width for any camera height.
• Lower cameras are generally less affected than higher ones. Areas during swing look most affected (distal limbs blocked by proximal limbs).
• For a balanced setup (i.e. all cameras or maybe just having cameras at multiple heights), there is almost no effect of width.
• Camera height adjusts the offset/starting point/average of the signals up or down, with low cameras having more ext. rot and high cameras having less.
• Looks like the leg axial rotation moves from being accounted for in the knee to the hip as the camera height increases. Results in opposite effects of height on those two joints.

4 Heights by 3 Widths

The plots summarize the effect of both camera height (group) as well as volume width. Data using all 32 cameras (Group0) is removed to only show data resulting from 8-camera setups. The augmented setups for width3 (Group51 and Group62) are also removed.

32 Camera Group at Each Width

These plots attempt to isolate purely the effect of width by using data from all cameras (group0) assumed to provide the best predictions.

Widest vs. Narrowest Widths (Width1 vs Width3)

These plots compare the widest (~6.9 m) and narrowest (3.3m) widths for each camera height. The augmented groupings for width3 are not included

Augmented Hallway-like Setups (Width 3)

These plots compare the narrowest (3.3m) width with the inclusion of the augmented groups51 and group62 that include saggital views that capture the full subject. The saggital views for Group 1 and Group 2 in width 3 do not get full body as they are too close.

The data from Group0 in width 1 is used a reference in black.

Width1 Only: 6.9m

Width2 Only: 5.5m

Width3 Only: 3.3m